主题: A Consistent Variable Screening Procedure for Classification

报告人: Prof. Xiaodan Fan, The Chinese University of Hong Kong

时间: 9:30 - 10:30, Tuesday, May 19, 2020

地点:  Zoom, Meeting ID: 559-916-3678 (Link: https://cuhksz.zoom.us/j/5599163678)

摘要：

In the era of big data, selecting important variables out of numerous candidates is often the first step and arguably the most important step for constructing good classifiers or regression models. Here we propose a new screening procedure for high-dimensional binary classification, which enjoys the consistent screening property instead of the weaker sure screening property of most existing methods. It is based on a non-parametric test statistic for the problem of two-sample distribution comparison. The test statistic is designed to combine the merits of the Chi-square statistic and the Kolmogorov-Smirnov statistic.  The screening procedure is much more powerful than other methods over a broad range of cases. Extensive simulations and an application on diabetes data showed the effectiveness and advantages of our new method.

简介：

Dr. Xiaodan Fan is an Associate Professor and the Graduate Division Head of Department of Statistics, The Chinese University of Hong Kong. He received his Ph.D. degree in Statistics from Harvard University. Before that, he got his B.E. degree in Automation and M.S. degree in Pattern Recognition & Intelligent Systems from Tsinghua University. Dr. Fan is interested in probabilistic modeling and statistical computing, especially for the problems from computational biology/bioinformatics.